Neural Network GPU training with output weights fails with "No method 'subsref' "

2 次查看（过去 30 天）

显示更早的评论

Matt Flax 2020-8-19

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/581582-neural-network-gpu-training-with-output-weights-fails-with-no-method-subsref

评论： Aditya Patil 2020-9-21

在 MATLAB Online 中打开

When training a NN I would like to use output weights (between 0 and 1), but I get the following error :

No method 'subsref' with matching signature found for class 'gpuArray'

My function does the following :

% get the gpuDevice
% load data
% change 'tansig' transferFcn to 'elliotsig'
for i=1:net.numLayers
  if strcmp(net.layers{i}.transferFcn,'tansig')
    net.layers{i}.transferFcn = 'elliotsig';
  end
end
% I had to alter processFcns to avoid the following errors :
% Input processing function REMOVECONSTANTROWS is not supported with GPU.
net.input.processFcns = {'mapminmax'};
net.output.processFcns = {'mapminmax'};
% convert data to GPU data with the nndata2gpu function
% Then train
net2 = train(net2, x, t, [], [], wtsOutG);

The error returned is :

No method 'subsref' with matching signature found for class 'gpuArray'

1 个评论
显示 -1更早的评论隐藏 -1更早的评论

Aditya Patil 2020-9-21

Can you provide a sample code snippet that creates the error? I am unable to reproduce to error with the given code, hence it might be possible that the issue lies in rest of the code.

请先登录，再进行评论。

请先登录，再回答此问题。